Dependency errors stemming from scikit-image+numpy

[oesteban]

@tsalo has hit the dependency issues we have been seeing in NiWorkflows and SDCFlows lately. Solution in nipreps/sdcflows#158.

[tsalo]

I'm a little confused by the issue and the fix. It looks like sdcflows' fix was Python <3.7-specific (i.e., just install numpy with no specific version mentioned if Python 3.7+), while fMRIPrep requires 3.7+, and already requires numpy without a specific version mentioned.

[effigies]

numpy <1.20; python_version < "3.7" should only be relevant if you're using python setup.py install, rather than pip install, which you shouldn't be doing. Is your situation different?

[tsalo]

I haven't tried anything locally- I'm just focused on the test failures in #2349 and #2366.

[effigies]

Ah, so it's a super-old numpy in the Docker image conflicting with a new scikit-image. I'll try upgrading everything to the latest release that isn't a x.y.0.

[tsalo]

Oh... so it's just tied to the Docker. That makes sense. Thanks!

[oesteban]

This should be resolved now.